The Market Making Book

14. Which Algorithm for Which Market

Sports, stocks, and crypto are three different physics. Matching the algorithm to the price process is half the job.

Part IV · Chapter 14
DimensionSports / predictionStocksCrypto
Price processJump-diffusion on [0,1], vol crush at resolutionDiffusion + scheduled jumps (earnings, macro)Diffusion + unscheduled jumps, 24/7
Fair-value anchorExternal model (Markov match model)Microprice + order-flow signalsCross-venue price + microprice
Dominant riskCourtsider adverse selection at eventsInstitutional informed flow; queue gamesCross-exchange arb pick-offs; trend inventory
HedgingEssentially none (binary, no underlying)Rich (correlated names, ETFs, options)Excellent (perps, other venues)
Fee regimeParabolic taker fees ⇒ maker-onlyRebate capture; reg constraintsTiered; rebates at volume
Best-fit algorithmsModel-anchored AS-in-logit + event gating (Prop. A/B)GLFT + microprice + queue craftCross-exchange hedged MM, grids in ranges, funding-aware perps MM

Reading the table strategically

  • Sports: your edge is the model and the event reflex, not latency-in-general. A correct fair value 300ms after a point beats a fast-but-naive quote every time. Unhedgeable inventory means caps must be honest.
  • Stocks: the rebate and the queue are the business; the math (GLFT) governs the residual inventory. For most independents, options spreads are the more realistic prize (Ch. 13).
  • Crypto: the hedge leg transforms the problem: with instant hedging, inventory risk shrinks and the game becomes fees + slippage + toxicity. Funding adds a carry dimension unique to perps. (Chapter 17 walks the complete build, from signal to live bot.)
One principle above allThe algorithm follows the price process; the price process follows the asset. Never port a quoting engine across asset classes without re-deriving what σ, jumps, (T−t), and "fair value" even mean there.

On this page

GitHubGitHub repository